05 - Application Layer

Problem 1: Short Answer Questions

• (a) The available bandwidth on transcontinental links is increasing every year, yet the round-trip latency is starting to approach the speed of light limits. What are the implications of this for applications such as HTTP?

  View Answer

  In short: The increasing bandwidth-delay product means applications become more latency-limited than bandwidth-limited. HTTP performance is increasingly constrained by round-trip time rather than download speed, making latency optimization (connection reuse, pipelining, multiplexing) more critical than ever.

  Elaboration:

  The Bandwidth-Delay Product Problem:

  Bandwidth Delay Product = Bandwidth × RTT
  
  Example:
  - 1 Gbps link, 100 ms RTT = 100 Mb = 12.5 MB
  - 10 Gbps link, 100 ms RTT = 1 Gb = 125 MB
  
  This represents "data in flight" on the link at any moment.
  

  What This Means for HTTP:

  1. Latency becomes the bottleneck

     Scenario: Download 1 MB file
     
     Slow link (10 Mbps, 50 ms RTT):
       Time = TCP handshake (50 ms) + HTTP request (50 ms) +
              transmission (800 ms) = ~900 ms
       Dominated by transmission time
     
     Fast link (1 Gbps, 100 ms RTT):
       Time = TCP handshake (100 ms) + HTTP request (100 ms) +
              transmission (8 ms) = ~208 ms
       Dominated by latency/roundtrips!
     

  2. Connection overhead matters more

     With HTTP/1.0 (new connection per object):
       For 10 objects: 10 × TCP handshake (3 × RTT) = 30 RTTs
     
     With fast link, 100 ms RTT:
       = 3 seconds just for handshakes!
     
     Transmission time for small objects becomes negligible
     

  3. Why bandwidth improvements help less

     Going from 100 Mbps to 1 Gbps (10× improvement):
       Saves ~8 ms per 1 MB
     
     But RTT is still 100 ms (physics limit)
       = 1200× slower than speed-of-light
     
     Latency is the real constraint
     

  Implications for HTTP Applications:

  • Connection Reuse Critical: Persistent connections (HTTP/1.1) become essential
  • Pipelining/Multiplexing Needed: HTTP/2, HTTP/3 multiplexing over single connections
  • DNS Caching Essential: DNS lookups add 50-200 ms per domain
  • Geographic Distribution: Content delivery networks (CDNs) needed to reduce latency
  • Protocol Overhead Matters: TCP/TLS handshakes become dominant cost
  • Parallel Connections Less Useful: Can’t pipeline effectively anyway due to RTT

  Real World Example:

  Downloading webpage with 50 objects from different CDNs:
  
  Old approach (HTTP/1.0, 6 parallel connections):
    50 objects × 100 ms RTT = 5 seconds
    Bandwidth barely used
  
  New approach (HTTP/2 multiplexing, single connection):
    ~200 ms (1-2 RTTs for connection + pipelining)
    Same bandwidth, much faster
  

  Conclusion:

  As bandwidth increases, applications become latency-bound rather than bandwidth-bound. HTTP design must minimize RTTs through connection multiplexing, protocol efficiency, and geographic proximity rather than expecting bandwidth improvements to help.

• (b) What’s an authoritative name server? What part of the name space hierarchy is City University of New York (CUNY) name server is responsible for. Briefly explain.

  View Answer

  In short: An authoritative name server is the official source for DNS records in a particular zone of the namespace hierarchy. CUNY’s name server (cuny.edu) is responsible for the cuny.edu domain zone, containing all hosts within CUNY (qc.cuny.edu, hunter.cuny.edu, etc.).

  Elaboration:

  What is an Authoritative Name Server?

  An authoritative name server:
  - Maintains the official DNS records for a specific zone
  - Responds to queries about hosts in that zone
  - Is the "source of truth" for that zone
  - Does NOT perform recursive queries (typically)
  

  DNS Hierarchy:

  Root zone (.)
  ├── .edu zone (root delegates)
  │   ├── .cuny.edu zone (edu delegates)
  │   ├── .mit.edu zone
  │   └── .stanford.edu zone
  │
  └── .com zone
      ├── .google.com zone
      └── .amazon.com zone
  

  CUNY’s Responsibility:

  CUNY's authoritative nameserver (ns.cuny.edu or ns1.cuny.edu):
  
  Responsible zone: cuny.edu
  
  Contains records for:
  - **(qc)**cuny.edu (Queens College)
  - **(hunter)**cuny.edu (Hunter College)
  - **(baruch)**cuny.edu (Baruch College)
  - **(host1)**cuny.edu
  - **(host2)**cuny.edu
  - ... any host *.cuny.edu
  
  Does NOT contain:
  - **(mit)**edu records (MIT's server handles)
  - **(google)**com records (Google's server handles)
  - Any hosts outside cuny.edu
  

  How it Works:

  Query: What is the IP of qc.cuny.edu?
  
  1. Client → Root nameserver: "Who handles .edu?"
     Root: "Ask ns.edu.cuny or similar"
  
  2. Client → .edu nameserver: "Who handles .cuny.edu?"
     .edu: "Ask ns.cuny.edu (CUNY's nameserver)"
  
  3. Client → ns.cuny.edu (authoritative): "IP of qc.cuny.edu?"
     CUNY: "It's 136.48.100.1" (authoritative answer)
  

  Key Characteristics:

  Aspect	Authoritative	Non-Authoritative (Resolver)
  Source	Official zone records	Cached records
  Updates	Maintained by zone admin	Cached with TTL
  Responsibility	One zone only	Multiple zones (via caching)
  Record Type	Complete for zone	Partial (what was requested before)

  Conclusion:

  CUNY’s authoritative nameserver manages the cuny.edu zone and knows the official IP addresses for all hosts within that zone (qc.cuny.edu, hunter.cuny.edu, etc.). It is the authoritative source for these records in the DNS hierarchy.

• (c) Suppose a user requests a Web page that consists of some text and two images. Will the client send one request and receive 3 response messages? Explain.

  View Answer

  In short: It depends on the HTTP version and connection type. With HTTP/1.0 or non-persistent connections, the answer is NO—the client sends 3 requests (one for HTML, one for each image) and receives 3 responses. With HTTP/1.1 persistent connections or HTTP/2, it’s more complex.

  Elaboration:

  HTTP/1.0 with Non-Persistent Connections:

  Web page structure:
  - HTML document (text)
  - Image 1
  - Image 2
  
  Client behavior:
  1. Send HTTP GET request for HTML
     Receive response #1 (HTML)
  
  2. Parse HTML, find <img> references
     Send HTTP GET request for Image 1
     Receive response #2 (Image 1)
  
  3. Send HTTP GET request for Image 2
     Receive response #3 (Image 2)
  
  Result: 3 requests, 3 responses ✓
  

  HTTP/1.1 with Persistent Connections:

  Client behavior:
  1. Send GET for HTML
     Receive response #1
  
  2. Same TCP connection still open!
     Send GET for Image 1
     Receive response #2
  
  3. Same TCP connection still open!
     Send GET for Image 2
     Receive response #3
  
  Result: 3 requests, 3 responses
  But: All over the SAME TCP connection (more efficient)
  

  HTTP/1.1 with Pipelining:

  Client behavior (if pipelining enabled):
  1. Send GET request for HTML
  2. Send GET request for Image 1 (without waiting for response)
  3. Send GET request for Image 2 (without waiting for response)
  
  Then receive:
  - Response #1 (HTML)
  - Response #2 (Image 1)
  - Response #3 (Image 2)
  
  Result: 3 requests, 3 responses
  But: Requests pipelined, responses arrive in order
  

  HTTP/2 with Multiplexing:

  Client behavior:
  1. Single TCP connection
  2. Send frame for HTML
  3. Send frame for Image 1
  4. Send frame for Image 2
  
  Server can interleave responses:
  - Send HTML chunks
  - Send Image 1 chunks
  - Send Image 2 chunks
  - All on same connection, simultaneously
  
  Result: 3 responses, but not necessarily discrete "messages"
  

  The Key Point:

  Aspect	HTTP/1.0	HTTP/1.1 Persistent	HTTP/2
  Requests	3 (separate connections)	3 (same connection)	3 (same connection)
  Responses	3 (separate)	3 (same connection)	3 (multiplexed)
  Sequential?	Yes	Yes (default)	No (interleaved)
  Efficiency	Poor	Good	Excellent

  Conclusion:

  The answer is technically YES (3 requests → 3 responses), but the nuance depends on HTTP version:

  • HTTP/1.0: Three separate TCP connections, three distinct responses
  • HTTP/1.1+: One persistent connection, three responses in order
  • HTTP/2: One connection, three responses multiplexed together

• (d) Can two distinct Web pages from the same origin server, e.g., www.mit.edu/research.html and www.mit.edu/students.html, be sent over the same persistent connection? Why or why not?

  View Answer

  In short: YES. HTTP/1.1 persistent connections allow multiple requests and responses to be exchanged over the same TCP connection. Two distinct web pages from the same origin server can absolutely be sent over the same persistent connection.

  Elaboration:

  How Persistent Connections Work:

  Traditional (HTTP/1.0, non-persistent):
  1. TCP connection established
  2. GET /research.html
  3. Receive response (research.html)
  4. TCP connection closed
  
  5. NEW TCP connection established
  6. GET /students.html
  7. Receive response (students.html)
  8. TCP connection closed
  

  With HTTP/1.1 Persistent Connection:

  1. TCP connection established (SYN, SYN-ACK, ACK)
  
  2. GET /research.html
     Receive response (research.html)
     Connection stays OPEN
  
  3. GET /students.html (same TCP connection!)
     Receive response (students.html)
     Connection stays OPEN
  
  4. TCP connection closed (when idle timeout or explicit close)
  

  Benefits:

  Benefit	Impact
  No TCP handshake overhead	Save 3 RTTs per page
  No SSL/TLS renegotiation	Save 2 RTTs if HTTPS
  Connection warm-up	Congestion window increases
  Network efficiency	Better link utilization

  Example Timeline:

  Time 0 ms:
    Send: GET /research.html
  
  Time 50 ms:
    Receive: 200 OK + research.html
  
  Time 60 ms:
    Send: GET /students.html (same connection)
  
  Time 110 ms:
    Receive: 200 OK + students.html
  
  Total time: 110 ms
  
  If separate connections:
    Connection 1: TCP handshake (50 ms) + request/response (50 ms) = 100 ms
    Connection 2: TCP handshake (50 ms) + request/response (50 ms) = 100 ms
    Total: 200 ms (90 ms extra!)
  

  HTTP Request Format (same connection):

  GET /research.html HTTP/1.1
  Host: www.mit.edu
  Connection: keep-alive
  
  [Server sends response, connection remains open]
  
  GET /students.html HTTP/1.1
  Host: www.mit.edu
  Connection: keep-alive
  
  [Server sends response, connection remains open]
  

  Conditions:

  Persistent connections work when:
  1. Both pages from SAME server (www.mit.edu)
  2. HTTP/1.1 is used (default in modern browsers)
  3. Connection header not set to "close"
  4. Content-Length or chunked encoding provided
  5. No HTTP errors that close connection (500, 503, etc.)
  

  Conclusion:

  YES, two distinct web pages from the same origin server can be sent over the same persistent connection. This is the default behavior in HTTP/1.1, and it significantly improves performance by eliminating TCP handshake overhead.

• (e) Can two distinct Web pages from different origin servers, e.g., www.mit.edu/research.html and www.cuny.edu/students.html, be sent over the same persistent connection? Why or why not?

  View Answer

  In short: NO. Persistent connections are specific to a single server. A connection to www.mit.edu cannot be reused for requests to www.cuny.edu. The client must establish a separate TCP connection to each origin server.

  Elaboration:

  Why Not?

  HTTP/1.1 persistent connections are tied to:
  1. Host (e.g., www.mit.edu)
  2. Port (e.g., 80 for HTTP)
  3. Protocol (HTTP vs HTTPS)
  
  Connection to www.mit.edu:80 is separate from www.cuny.edu:80
  Cannot be reused across different servers
  

  TCP Connection Mechanics:

  TCP connection identified by 5-tuple:
  - Source IP
  - Source port
  - Destination IP (www.mit.edu = 128.30.2.36)
  - Destination port (80)
  - Protocol (TCP)
  
  Connection to www.cuny.edu (136.48.0.1) would be:
  - Source IP (same)
  - Source port (different)
  - Destination IP (different!) ← DIFFERENT SERVER
  - Destination port (80)
  - Protocol (TCP)
  
  Completely different connection
  

  HTTP Request Format (different servers):

  ← Connection to www.mit.edu
  GET /research.html HTTP/1.1
  Host: www.mit.edu
  
  [Response received]
  [Connection closed or kept open for more mit.edu requests]
  
  ← NEW Connection to www.cuny.edu
  GET /students.html HTTP/1.1
  Host: www.cuny.edu
  
  [Response received]
  

  Timeline Comparison:

  Same server (www.mit.edu):
  Time 0: GET /research.html
  Time 50: Receive response
  Time 60: GET /students2.html (same TCP connection)
  Time 110: Receive response
  Total: 110 ms
  
  Different servers (www.mit.edu vs www.cuny.edu):
  Time 0: TCP handshake to mit.edu
  Time 50: GET /research.html
  Time 100: Receive response
  Time 101: TCP handshake to cuny.edu (NEW connection)
  Time 151: GET /students.html
  Time 201: Receive response
  Total: 201 ms (extra TCP handshake!)
  

  Exception: HTTP Proxies

  A proxy can maintain persistent connections to multiple servers:
  
  Browser → Proxy: GET www.mit.edu/research.html
  Proxy ← → www.mit.edu (connection 1)
  
  Browser → Proxy: GET www.cuny.edu/students.html
  Proxy ← → www.cuny.edu (connection 2)
  
  But the browser itself still only connects to ONE proxy
  Proxy manages connections to multiple servers
  

  Modern Workaround: CDNs

  Instead of different servers:
  - Both pages served from CDN edge server
  - Same origin server (CDN node)
  - Persistent connection works
  
  Browser → CDN node for mit.edu content
  Browser → CDN node for cuny.edu content (same CDN server)
  Can reuse connection within CDN
  

  Conclusion:

  NO, two web pages from different origin servers cannot use the same persistent connection. Each server requires a separate TCP connection. This is a fundamental limitation of TCP (which is server-specific) and HTTP (which respects TCP connection boundaries).

• (f) With nonpersistent connections between the browser and the origin server, is it possible for a single TCP segment to carry two distinct HTTP request messages. Explain.

  View Answer

  In short: NO. With nonpersistent connections, each HTTP request requires its own TCP connection (3-way handshake, request, response, close). A TCP segment carries data from one connection only, so two HTTP requests would require two separate TCP connections and thus two separate segments.

  Elaboration:

  Understanding TCP Segments:

  A TCP segment is the unit of data at the transport layer
  - Contains TCP header + payload (HTTP data)
  - Belongs to ONE TCP connection (identified by source/dest IP:port)
  
  One segment = One TCP connection
  Cannot carry data from two different connections
  

  Nonpersistent Connection Model:

  Request 1:
  1. SYN (TCP handshake)
  2. SYN-ACK
  3. ACK
  4. [TCP segment with HTTP GET request]
     Carries: GET /page1.html HTTP/1.0\r\n...
  
  5. [Response received, connection closes]
  
  Request 2:
  6. NEW SYN (new TCP connection)
  7. SYN-ACK
  8. ACK
  9. [NEW TCP segment with HTTP GET request]
     Carries: GET /page2.html HTTP/1.0\r\n...
  
  10. [Response received, connection closes]
  

  Why Not in One Segment?

  Hypothesis: Send both in one segment?
  
  GET /page1.html HTTP/1.0\r\n...
  GET /page2.html HTTP/1.0\r\n...
  
  Problem 1: Which connection?
  - TCP connection is between specific IP:port pairs
  - One connection to server A
  - One connection to server B (different)
  - Can't fit both in one segment
  
  Problem 2: HTTP protocol expectation
  - Server receives segment on established connection
  - Reads first request: GET /page1.html
  - Sends response for page1
  - Connection closes (nonpersistent)
  - Second request is lost!
  
  Problem 3: Multiple requests aren't delimited
  - How would server know where one HTTP message ends?
  - Without Content-Length or keep-alive, message ends with connection close
  

  What WOULD Work (but breaks nonpersistent model):

  If we COULD send two requests in one segment:
  
  GET /page1.html HTTP/1.1\r\n
  Host: server.com\r\n
  Connection: keep-alive\r\n
  Content-Length: 0\r\n
  \r\n
  GET /page2.html HTTP/1.1\r\n
  Host: server.com\r\n
  Content-Length: 0\r\n
  \r\n
  
  But this REQUIRES:
  - Persistent connection (HTTP/1.1)
  - Keep-alive header
  - Proper message framing
  - This is NOT nonpersistent!
  

  TCP and HTTP Constraints:

  Constraint	Implication
  Nonpersistent = new TCP connection per request	Each request needs own 3-way handshake
  TCP segment belongs to one connection	Can’t mix requests from different connections
  Connection close ends message	Second request lost when connection closes
  HTTP/1.0 (nonpersistent) has no framing	Can’t delimit multiple requests

  Example Timeline:

  Time 0:   SYN ——————→
  Time 10:  ←———— SYN-ACK
  Time 20:  ACK ——————→
  Time 30:  GET request ——→ [One segment carries one HTTP request]
  Time 80:  ←———— Response
  Time 90:  FIN ——————→ [Connection closes]
  
  Time 91:  NEW SYN ———→ [New connection for second request]
  Time 101: ←——— SYN-ACK
  Time 111: ACK ——————→
  Time 121: GET request ——→ [Different segment, different connection]
  Time 171: ←———— Response
  

  Conclusion:

  NO. A TCP segment cannot carry two distinct HTTP request messages in a nonpersistent connection model because:

  1. Each request requires its own TCP connection
  2. A TCP segment belongs to exactly one connection
  3. Nonpersistent connections close after response, losing any additional data
  4. HTTP/1.0 has no framing mechanism to delimit multiple messages

  This is why persistent connections (HTTP/1.1) were invented—to allow multiple requests over one connection and better utilize bandwidth.

• (g) We know that a separate TCP connection is established for data transfer in FTP. Briefly describe the client and server communication to make this possible.

  View Answer

  In short: FTP uses two TCP connections: a control connection for commands and a data connection for file transfer. The client sends commands (USER, PASS, RETR, STOR) over the control connection, and the server establishes a data connection when needed, either in active mode (server initiates) or passive mode (client initiates).

  Elaboration:

  Two Connection Model:

  FTP Client                          FTP Server
  
  Control connection ←———————————→ Control port 21
  (client commands,
   server responses)
  
  Data connection ←———————————→ Data port 20 (active)
                                or random port (passive)
  (file data transfer)
  

  Active Mode (Server-Initiated Data Connection):

  Step 1: Control Connection Established
    Client → Server: TCP connection to port 21
    This persists for entire FTP session
  
  Step 2: User Authentication
    Client → Server (control): USER username\r\n
    Server → Client (control): 331 Password required\r\n
    Client → Server (control): PASS password\r\n
    Server → Client (control): 230 Login successful\r\n
  
  Step 3: Issue Retrieve Command
    Client → Server (control): RETR filename\r\n
    Server → Client (control): 150 Opening data connection\r\n
  
  Step 4: Server Initiates Data Connection
    Server → Client: TCP connection from port 20 to client's data port
    [Data transfer begins on this connection]
    [Transfer completes]
    Server → Client (data): Connection closes
  
  Step 5: Server Notifies Completion
    Server → Client (control): 226 Transfer complete\r\n
  

  Passive Mode (Client-Initiated Data Connection):

  Motivation: Firewalls/NAT often block incoming connections
  
  Step 1-2: Control Connection & Authentication (same as active)
  
  Step 3: Request Passive Mode
    Client → Server (control): PASV\r\n
    Server → Client (control): 227 Entering Passive Mode (h1,h2,h3,h4,p1,p2)\r\n
    [Response contains server's IP and random port number]
  
  Step 4: Client Initiates Data Connection
    Client → Server: TCP connection to provided IP:port
    (Server was listening on this port)
  
  Step 5: Issue Retrieve Command
    Client → Server (control): RETR filename\r\n
    Server → Client (control): 150 Opening data connection\r\n
    [Data transfer on already-established data connection]
    [Transfer completes]
    Data connection closes
  
  Step 6: Completion Notification
    Server → Client (control): 226 Transfer complete\r\n
  

  Active Mode Timeline:

  Time 0:    Control connection (client port → server:21)
  Time 10:   USER command sent
  Time 20:   PASS command sent
  Time 30:   RETR filename command sent
  Time 40:   ← Server initiates data connection (server:20 → client port X)
  Time 50:   Data transfer begins
  Time 500:  Data transfer completes
  Time 510:  ← 226 Transfer complete (control connection)
  

  Passive Mode Timeline:

  Time 0:    Control connection (client port → server:21)
  Time 10:   USER command sent
  Time 20:   PASS command sent
  Time 30:   PASV command sent
  Time 40:   ← Server responds with port number (e.g., 1234)
  Time 50:   Data connection initiated (client → server:1234)
  Time 60:   RETR filename command sent
  Time 70:   Data transfer begins
  Time 500:  Data transfer completes
  Time 510:  ← 226 Transfer complete (control connection)
  

  Command Examples on Control Connection:

  Command	Purpose	Response
  USER	Provide username	331 (need password)
  PASS	Provide password	230 (success) or 530 (fail)
  RETR	Retrieve file	150 (opening data), 226 (done)
  STOR	Store file	150 (opening data), 226 (done)
  LIST	List directory	150 (opening data), 226 (done)
  QUIT	End session	221 (goodbye)

  Why Separate Data Connection?

  1. Protocol separation
     - Control: Command/response (ASCII text, small)
     - Data: File transfer (binary, large)
  
  2. Flexibility
     - Can transfer multiple files without re-authenticating
     - Can use different data rates
     - Can resume interrupted transfers
  
  3. Network efficiency
     - Control connection lightweight
     - Data connection optimized for throughput
  
  4. Compatibility
     - Works with firewall rules
     - Can use active or passive depending on network
  

  Key Points:

  - Control connection: Always client → server:21 (persists)
  - Data connection: Separate, established per transfer
  - Active: Server initiates data connection from port 20
  - Passive: Client initiates to server's random port
  - Responses on control connection indicate data connection status
  

  Conclusion:

  FTP uses a control connection (to port 21) for commands and responses, and a separate data connection (port 20 for active, random port for passive) for actual file transfer. The client sends FTP commands over the control connection, and the server initiates (active mode) or accepts (passive mode) the data connection as needed. This separation allows efficient file transfer while maintaining session control and enabling features like authentication and error reporting.

• (h) Does a user e-mail agent upload an outgoing e-mail to a mail server using POP3/IMAP, or SMTP? Briefly explain.

  View Answer

  In short: SMTP is used to upload outgoing email. POP3 and IMAP are used only for downloading/retrieving received email. The mail client uses SMTP to send messages to the mail server, which then routes them to recipients.

  Elaboration:

  Three Separate Protocols:

  User's Mail Client
  ↓
  SMTP (port 25, 465, 587): SEND outgoing email
  ↓
  Mail Server (SMTP server)
  ↓
  Routes to recipient's mail server via SMTP
  ↓
  Recipient's Mail Server (POP3/IMAP server)
  ↓
  POP3 or IMAP (port 110, 143, 993): RETRIEVE email
  ↓
  Recipient's Mail Client
  

  SMTP (Simple Mail Transfer Protocol):

  Purpose: SENDING email
  
  Client workflow:
  1. User composes email in mail client (Outlook, Gmail, etc.)
  2. User clicks "Send"
  3. Mail client connects to SMTP server (port 587 with TLS)
  4. Client authenticates: AUTH LOGIN
  5. Client sends message:
     - MAIL FROM: sender@domain.com
     - RCPT TO: recipient@domain.com
     - DATA (message body)
  6. Server accepts and routes message
  7. Connection closes
  
  Server then:
  - Looks up recipient's mail server via DNS
  - Connects to recipient's SMTP server
  - Delivers message
  - (May queue if recipient offline)
  

  POP3 (Post Office Protocol 3):

  Purpose: RETRIEVING email (download-and-delete model)
  
  Client workflow:
  1. Mail client connects to POP3 server (port 110 or 995)
  2. User authenticates with username/password
  3. Server returns list of messages
  4. Client downloads messages
  5. Messages deleted from server (typically)
  6. Connection closes
  
  Characteristic:
  - Intended for single client access
  - After download, email usually removed from server
  - Not ideal for multiple devices
  

  IMAP (Internet Message Access Protocol):

  Purpose: RETRIEVING email (keep-on-server model)
  
  Client workflow:
  1. Mail client connects to IMAP server (port 143 or 993)
  2. User authenticates
  3. Server provides folder structure (Inbox, Drafts, Sent, etc.)
  4. Client can:
     - Preview messages without downloading
     - Download specific messages
     - Delete, flag, organize messages
     - Synchronize across devices
  5. Messages stay on server
  6. Connection can remain open
  
  Characteristic:
  - Designed for multiple client access
  - Emails remain on server until explicitly deleted
  - Great for accessing from multiple devices
  - More bandwidth-efficient (selective download)
  

  Complete Email Flow:

  Alice sends email to Bob:
  
  Step 1 (SMTP - Send):
    Alice's client → SMTP server (mail.alice.com:587)
    Sends: alice@alice.com → bob@bob.com
  
  Step 2 (SMTP - Route):
    mail.alice.com → mail.bob.com (SMTP)
    Message transferred between servers
  
  Step 3 (IMAP/POP3 - Receive):
    Bob's client → mail.bob.com (IMAP port 993)
    Bob downloads/reads message
  

  Key Distinction:

  Protocol	Direction	Purpose	Port
  SMTP	Client → Server	SEND email	25, 465, 587
  POP3	Client ← Server	RETRIEVE email	110, 995
  IMAP	Client ← Server	RETRIEVE email	143, 993

  Conclusion:

  SMTP is used for uploading/sending outgoing email to the mail server. POP3 and IMAP are used for downloading received email from the mail server. These are distinct protocols with different purposes in the email infrastructure.

• (i) What’s a MIME type? What is it used for? Briefly explain.

  View Answer

  In short: A MIME type (Multipurpose Internet Mail Extensions) is a standard label that identifies the format of data (e.g., text/plain, image/jpeg, application/pdf). It tells systems how to interpret and display the content, enabling proper handling of diverse file types across networks.

  Elaboration:

  What is MIME?

  MIME Type Syntax:
  type/subtype
  
  Examples:
  - text/plain (plain text)
  - text/html (HTML document)
  - image/jpeg (JPEG image)
  - image/png (PNG image)
  - application/pdf (PDF document)
  - application/json (JSON data)
  - audio/mpeg (MP3 audio)
  - video/mp4 (MP4 video)
  - application/zip (ZIP archive)
  

  Purpose:

  Without MIME types:
  - System receives file called "document"
  - Is it text? Binary? Image? Archive?
  - How should it be displayed?
  - What program should open it?
  - Ambiguous and error-prone
  
  With MIME types:
  - Server sends "Content-Type: application/pdf"
  - Client knows it's a PDF
  - Client launches PDF reader
  - Content displayed correctly
  

  Common MIME Types:

  Type	Subtype	Purpose
  text	plain, html, css, javascript	Text-based files
  image	jpeg, png, gif, svg+xml	Image files
  audio	mpeg, wav, ogg	Audio files
  video	mp4, webm, ogg	Video files
  application	pdf, json, xml, zip, octet-stream	Data/binary files
  multipart	form-data, mixed, related	Multiple parts in one message

  How MIME Works in HTTP:

  HTTP Response:
  HTTP/1.1 200 OK
  Content-Type: text/html; charset=utf-8
  Content-Length: 1234
  
  <!DOCTYPE html>
  <html>
  ...
  

  Browser sees “text/html” → Renders as HTML webpage

  HTTP Response:
  HTTP/1.1 200 OK
  Content-Type: application/pdf
  Content-Length: 50000
  
  [binary PDF data]
  

  Browser sees “application/pdf” → Launches PDF viewer

  MIME in Email:

  Email with attachment:
  
  From: alice@example.com
  To: bob@example.com
  Subject: Photos
  MIME-Version: 1.0
  Content-Type: multipart/mixed
  
  --boundary123
  Content-Type: text/plain
  
  Here are the photos you requested.
  
  --boundary123
  Content-Type: image/jpeg
  Content-Transfer-Encoding: base64
  
  [binary image data encoded as base64]
  
  --boundary123--
  

  Mail client:

  • Reads “multipart/mixed”
  • Recognizes multiple parts
  • Displays text portion
  • Saves image/jpeg as attachment with .jpg extension

  MIME with Parameters:

  Content-Type: text/plain; charset=utf-8
  └─ Type: text
  ├─ Subtype: plain
  └─ Parameter: charset=utf-8 (UTF-8 encoding)
  
  Content-Type: image/jpeg; name="photo.jpg"
  └─ Type: image
  ├─ Subtype: jpeg
  └─ Parameter: filename for download
  
  Content-Type: multipart/form-data; boundary=----WebKitFormBoundary
  └─ Type: multipart
  ├─ Subtype: form-data
  └─ Parameter: boundary delimiter for parts
  

  Why MIME Matters:

  1. Content Negotiation

     • Server can offer multiple formats
     • Client requests preferred format
     • “Accept: text/html, application/json”

  2. Charset Handling

     • “text/html; charset=utf-8”
     • Ensures proper character encoding
     • Prevents garbled text

  3. Plugin/Handler Selection

     • OS looks at MIME type
     • Launches appropriate application
     • User doesn’t need to specify

  4. Interoperability

     • Standard way to describe content
     • Works across all platforms
     • Enables automation

  Conclusion:

  A MIME type is a standard label (e.g., text/html, image/jpeg) that identifies the format of data. It’s used to tell systems how to interpret, display, and handle content, enabling proper routing and processing of diverse file types across email systems, web servers, and applications.

• (j) Why might a domain address (e.g., www.cnn.com) have several IP addresses?

  View Answer

  In short: A domain can have multiple IP addresses for load balancing (distributing traffic across servers), geographic redundancy (serving from multiple locations), fault tolerance (if one server fails, others handle traffic), and scalability (handling large traffic volumes).

  Elaboration:

  Load Balancing:

  Single IP address (poor):
    All requests → Single server
    Server capacity: 1000 requests/sec
    If 2000 requests arrive: 50% get dropped
  
  Multiple IP addresses (good):
    DNS returns multiple IPs in rotation
    Requests distributed across servers
    Total capacity: 4000 requests/sec (4 servers × 1000 each)
  
  www.cnn.com might have:
  - **(93)**184.216.34
  - **(93)**184.216.35
  - **(93)**184.216.36
  - **(93)**184.216.37
  

  How DNS Round-Robin Works:

  Client 1 → DNS: What's the IP for www.cnn.com?
            ← DNS: [93.184.216.34, 93.184.216.35, 93.184.216.36, ...]
            Client gets first IP: 93.184.216.34
  
  Client 2 → DNS: What's the IP for www.cnn.com?
            ← DNS: [93.184.216.35, 93.184.216.36, ..., 93.184.216.34]
            (rotated list)
            Client gets: 93.184.216.35
  
  Client 3 → DNS: What's the IP for www.cnn.com?
            ← DNS: [93.184.216.36, ..., 93.184.216.34, 93.184.216.35]
            Client gets: 93.184.216.36
  
  Result: Traffic spread across all servers
  

  Geographic Redundancy:

  Single server (bad):
    Server in New York
    Los Angeles users: ~3000 ms latency
    Europe users: ~10000 ms latency
    User experience: Poor
  
  Multiple geographic locations (good):
    Server 1: New York (1.2.3.4)
    Server 2: Los Angeles (1.2.3.5)
    Server 3: London (1.2.3.6)
    Server 4: Tokyo (1.2.3.7)
  
    User in LA → Connects to 1.2.3.5 (local server)
    Latency: ~10 ms (much better)
  
  GeoDNS can be used:
    Clients in US get US servers
    Clients in Europe get EU servers
    Clients in Asia get Asia servers
  

  Fault Tolerance:

  Single server (risky):
    www.cnn.com → 1.2.3.4
    Server 1.2.3.4 crashes
    Website is DOWN
    All users affected
    No redundancy
  
  Multiple servers (safe):
    www.cnn.com → [1.2.3.4, 1.2.3.5, 1.2.3.6, 1.2.3.7]
  
    Server 1.2.3.4 crashes:
    Clients keep connecting to other IPs
    Users redirected automatically
    Website stays UP
    Minimal disruption
  
  Health checks:
    Monitoring service checks each server
    If server unhealthy: Remove from DNS responses
    Clients automatically avoid failed server
  

  Traffic Scalability:

  Peak traffic analysis:
  - Typical traffic: 10,000 requests/sec per server
  - Peak traffic (election day, breaking news): 100,000 requests/sec
  
  Single server solution:
    Would need 10 servers during peak only
    Expensive and wasteful
  
  Multiple permanent servers:
    Run 4-5 servers normally
    During peak: Some requests queued briefly
    Costs less than scaling to 10
    Handles most spikes gracefully
  

  Real Example: CNN

  www.cnn.com DNS lookup returns:
  
  ; <<>> dig www.cnn.com
  
  www.cnn.com. 300 IN A 151.101.1.67
  www.cnn.com. 300 IN A 151.101.65.67
  www.cnn.com. 300 IN A 151.101.129.67
  www.cnn.com. 300 IN A 151.101.193.67
  
  Note: These are Fastly CDN IPs in different geographic regions
  

  Other Reasons:

  1. CDN (Content Delivery Network)

     • Multiple edge servers worldwide
     • Each has own IP
     • Users served from nearest edge
     • Faster content delivery

  2. A/B Testing

     • Version A served from IP 1.2.3.4
     • Version B served from IP 1.2.3.5
     • Different users test different versions

  3. Graceful Degradation

     • During maintenance: Reduce IPs in DNS response
     • Gradually drain traffic from server being updated
     • Zero downtime deploys

  4. DDoS Mitigation

     • Multiple IPs spread attack traffic
     • Easier to filter/block attack sources
     • Continues serving through attack

  Conclusion:

  A domain has multiple IP addresses primarily for:

  • Load balancing (distribute traffic)
  • Geographic redundancy (serve from multiple locations)
  • Fault tolerance (survive server failures)
  • Scalability (handle traffic spikes)
  • Performance (users connect to nearest server)

  This is achieved through DNS round-robin, GeoDNS, health checks, and CDN architecture.

• (k) Why might a Web server have several IP addresses for a single interface?

  View Answer

  In short: A web server might have multiple IP addresses on a single network interface to host multiple domains/websites, serve different services on different IPs, implement virtual hosting, isolate traffic for different customers, or handle SSL/TLS certificates for multiple domains.

  Elaboration:

  Virtual Hosting:

  Single server, multiple websites:
  
  IP Configuration:
  eth0: 1.2.3.4 (www.site1.com)
  eth0: 1.2.3.5 (www.site2.com)
  eth0: 1.2.3.6 (www.site3.com)
  
  All on same physical network interface (eth0)
  All running on same server machine
  
  When client connects:
    Client → 1.2.3.4 (gets site1)
    Client → 1.2.3.5 (gets site2)
    Client → 1.2.3.6 (gets site3)
  

  Before SNI (Server Name Indication):

  SSL/TLS required different IP per domain:
  
  Problem: How does server know which cert to use?
  - HTTPS handshake happens before HTTP Host header
  - Server doesn't know which domain client wants
  - Can't select correct certificate
  
  Solution: One IP per SSL domain
  - **(www)**site1.com → IP 1.2.3.4 with Site1 cert
  - **(www)**site2.com → IP 1.2.3.5 with Site2 cert
  - **(www)**site3.com → IP 1.2.3.6 with Site3 cert
  
  Now server can identify domain from incoming IP
  

  Modern Era (with SNI):

  SNI (Server Name Indication) - TLS extension:
  - Client sends hostname during TLS handshake
  - Server knows which cert to use
  - Multiple domains can share one IP!
  
  But legacy support may still require:
  - Multiple IPs for older clients
  - Fallback for incompatible browsers
  

  Practical Scenarios:

  Scenario 1: Shared Hosting Provider

  Company: "WebHost.com" shared hosting
  
  One physical server hosts 100 customer websites:
  192.168.1.100:
    - IP address 203.0.113.1 → customer1.com
    - IP address 203.0.113.2 → customer2.com
    - IP address 203.0.113.3 → customer3.com
    - ... up to 203.0.113.100 → customer100.com
  
  Benefits:
  - Single server, multiple paying customers
  - Each customer has own IP (feels exclusive)
  - Different SSL certs for each
  

  Scenario 2: Service Isolation

  Large enterprise server configuration:
  
  eth0 (single physical NIC):
    - **(10)**0.1.10: Public-facing web server (www.company.com)
    - **(10)**0.1.11: Admin dashboard (secure, restricted access)
    - **(10)**0.1.12: API server (api.company.com)
    - **(10)**0.1.13: Backup/Health check IP
  
  Benefits:
  - Can apply different firewall rules per IP
  - Different QoS (Quality of Service) per IP
  - Easier to limit access (block 10.0.1.11 from outside)
  

  Scenario 3: Multi-Tenant Application

  SaaS platform: Multiple customers on same server
  
  eth0:
    - **(1)**2.3.100: Company A instance
    - **(1)**2.3.101: Company B instance
    - **(1)**2.3.102: Company C instance
  
  Each customer accesses their own IP:
    Company A employees → https://app.companyA.com (→ 1.2.3.100)
    Company B employees → https://app.companyB.com (→ 1.2.3.101)
    Company C employees → https://app.companyC.com (→ 1.2.3.102)
  
  Benefits:
  - Logical separation (feels like dedicated server)
  - Different SLA/performance tiers per customer
  - Can restart one customer's instance without affecting others
  

  Linux Configuration Example:

  # Configure multiple IPs on single interface (eth0)
  
  ip addr add 1.2.3.4/24 dev eth0
  ip addr add 1.2.3.5/24 dev eth0
  ip addr add 1.2.3.6/24 dev eth0
  
  # Verify:
  ip addr show eth0
  
  1: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP>
    inet 1.2.3.4/24 scope global eth0
    inet 1.2.3.5/24 scope global secondary eth0
    inet 1.2.3.6/24 scope global secondary eth0
  

  Web Server Configuration (Apache):

  # Virtual host on different IPs
  
  <VirtualHost 1.2.3.4:443>
    ServerName www.site1.com
    DocumentRoot /var/www/site1
    SSLCertificateFile /path/to/site1.crt
  </VirtualHost>
  
  <VirtualHost 1.2.3.5:443>
    ServerName www.site2.com
    DocumentRoot /var/www/site2
    SSLCertificateFile /path/to/site2.crt
  </VirtualHost>
  
  <VirtualHost 1.2.3.6:443>
    ServerName www.site3.com
    DocumentRoot /var/www/site3
    SSLCertificateFile /path/to/site3.crt
  </VirtualHost>
  

  Benefits Summary:

  Benefit	Use Case
  Virtual Hosting	Host multiple websites on one server
  SSL/TLS per domain	Each domain with own certificate (pre-SNI)
  Service Isolation	Different firewall rules per service
  Tenant Separation	Multi-tenant SaaS platforms
  Performance Control	Different rate limits per IP
  Billing/Accounting	Track usage per customer IP

  Modern Alternative: Name-Based Hosting

  With SNI support, can now use:
  
  eth0: 1.2.3.4 (only one IP)
  
  VirtualHosts:
  <VirtualHost 1.2.3.4:443>
    ServerName www.site1.com
    SSLEngine on
    SSLCertificateFile /path/to/site1.crt
  </VirtualHost>
  
  <VirtualHost 1.2.3.4:443>
    ServerName www.site2.com
    SSLEngine on
    SSLCertificateFile /path/to/site2.crt
  </VirtualHost>
  
  Client sends "ServerName" in TLS handshake
  Server picks correct cert based on name
  Multiple sites on single IP!
  

  Conclusion:

  A web server may have multiple IP addresses on a single interface for virtual hosting (multiple domains), SSL/TLS per domain (pre-SNI era), service isolation (different traffic types), tenant separation (multi-tenant platforms), or performance/billing management. While modern SNI enables multiple sites on one IP, multiple IPs may still be used for security, isolation, legacy compatibility, or administrative control.

• (l) Briefly describe what HEAD, GET, POST, PUT, PATCH, DELETE HTTP requests are used for.

  View Answer

  In short: GET retrieves data; POST sends data for server processing; HEAD is like GET but without response body; PUT replaces an entire resource; PATCH partially updates a resource; DELETE removes a resource. These form the foundation of RESTful APIs.

  Elaboration:

  GET - Retrieve Data

  Purpose: Request data without modifying server state
  
  Example:
  GET /api/users/123 HTTP/1.1
  Host: api.example.com
  
  Response:
  200 OK
  {
    "id": 123,
    "name": "John Doe",
    "email": "john@example.com"
  }
  
  Characteristics:
  - Data in URL query string: GET /users?id=123&sort=name
  - Idempotent (multiple identical requests = same result)
  - Safe (doesn't modify server state)
  - Cacheable (browsers cache GET responses)
  - Bookmarkable
  - Should NOT have request body
  

  POST - Create or Process Data

  Purpose: Submit data to server for processing (create, process, etc.)
  
  Example 1: Create new user
  POST /api/users HTTP/1.1
  Host: api.example.com
  Content-Type: application/json
  
  {
    "name": "Jane Doe",
    "email": "jane@example.com"
  }
  
  Response:
  201 Created
  Location: /api/users/124
  {
    "id": 124,
    "name": "Jane Doe",
    "email": "jane@example.com"
  }
  
  Example 2: Form submission
  POST /login HTTP/1.1
  Host: example.com
  Content-Type: application/x-www-form-urlencoded
  
  username=alice&password=secret123
  
  Characteristics:
  - Data in request body (hidden from URL)
  - NOT idempotent (repeated requests create multiple resources)
  - NOT safe (modifies server state)
  - Not cached (usually)
  - Not bookmarkable
  

  HEAD - Retrieve Headers Only

  Purpose: Like GET but without response body (just headers)
  
  Example:
  HEAD /document.pdf HTTP/1.1
  Host: example.com
  
  Response:
  200 OK
  Content-Type: application/pdf
  Content-Length: 50000
  Last-Modified: Mon, 01 Jan 2024 10:00:00 GMT
  
  (no body, but headers tell us file info)
  
  Use Cases:
  1. Check if resource exists without downloading
  2. Check file size before downloading
  3. Check last modification date
  4. Verify URL validity
  5. Bandwidth-efficient checks
  
  Characteristics:
  - Same as GET but no response body
  - Faster (no data transfer)
  - Useful for large files
  - Idempotent and safe
  

  PUT - Replace Entire Resource

  Purpose: Replace a resource entirely with new data
  
  Example: Update user 123 completely
  PUT /api/users/123 HTTP/1.1
  Host: api.example.com
  Content-Type: application/json
  
  {
    "name": "John Smith",
    "email": "john.smith@example.com",
    "phone": "555-1234"
  }
  
  Response:
  200 OK
  {
    "id": 123,
    "name": "John Smith",
    "email": "john.smith@example.com",
    "phone": "555-1234"
  }
  
  Key Difference (PUT vs POST):
  - PUT: Client specifies resource ID (PUT /users/123)
  - POST: Server generates resource ID (POST /users)
  
  Characteristics:
  - Replaces entire resource
  - Client specifies ID in URL
  - Idempotent (PUT twice = same result)
  - If resource doesn't exist: May create it (201) or error (404)
  

  PATCH - Partial Update

  Purpose: Partially update a resource (only changed fields)
  
  Example: Update only name field
  PATCH /api/users/123 HTTP/1.1
  Host: api.example.com
  Content-Type: application/json
  
  {
    "name": "John Smith"
  }
  
  Current state before PATCH:
  {
    "id": 123,
    "name": "John Doe",
    "email": "john@example.com",
    "phone": "555-0000"
  }
  
  Response (after PATCH):
  200 OK
  {
    "id": 123,
    "name": "John Smith",           ← Changed
    "email": "john@example.com",    ← Unchanged
    "phone": "555-0000"             ← Unchanged
  }
  
  PUT (for comparison - replaces all):
  PUT /api/users/123
  { "name": "John Smith" }
  
  Result with PUT:
  {
    "id": 123,
    "name": "John Smith",
    "email": null,        ← Lost!
    "phone": null         ← Lost!
  }
  (All fields not specified are removed/nulled)
  
  Characteristics:
  - Only changed fields required
  - More efficient than PUT
  - Idempotent (usually)
  - Not all servers support PATCH
  

  DELETE - Remove Resource

  Purpose: Delete a resource from server
  
  Example: Delete user 123
  DELETE /api/users/123 HTTP/1.1
  Host: api.example.com
  
  Response:
  204 No Content
  (Resource deleted, no body needed)
  
  Or:
  Response:
  200 OK
  { "message": "User 123 deleted successfully" }
  
  Characteristics:
  - Removes resource
  - Idempotent (deleting twice has same effect)
  - Safe to call multiple times
  - No request body (usually)
  - May return 404 if already deleted
  

  Summary Comparison:

  Method	Purpose	Idempotent	Safe	Body
  GET	Retrieve data	Yes	Yes	No
  HEAD	Retrieve headers	Yes	Yes	No
  POST	Create/process	No	No	Yes
  PUT	Replace resource	Yes	No	Yes
  PATCH	Partial update	Yes*	No	Yes
  DELETE	Remove resource	Yes	No	No

  REST API Example:

  Resource: /api/articles/42
  
  GET /api/articles/42
    → Retrieve article 42
  
  POST /api/articles
    → Create new article
  
  PUT /api/articles/42
    → Replace entire article 42
  
  PATCH /api/articles/42
    → Update some fields of article 42
  
  DELETE /api/articles/42
    → Delete article 42
  
  GET /api/articles
    → List all articles
  

  Conclusion:

  • GET: Read data (safe, idempotent)
  • HEAD: Check headers without body (efficient read)
  • POST: Create new resource or trigger action (not idempotent)
  • PUT: Replace entire resource (idempotent, client-specified ID)
  • PATCH: Partially update resource (idempotent, efficient update)
  • DELETE: Remove resource (idempotent)

  These form the CRUD operations (Create, Read, Update, Delete) for RESTful APIs.

• (m) Briefly describe the motivation for HTTP/2.0.

  View Answer

  In short: HTTP/2.0 was motivated by the need to reduce latency and overhead in HTTP/1.1, which suffered from head-of-line blocking, multiple connection limitations, and inefficient header transmission. HTTP/2 introduced multiplexing, binary framing, and header compression to dramatically improve performance.

  Elaboration:

  Problems with HTTP/1.1:

  Problem 1: Head-of-Line Blocking

  HTTP/1.1 limitation:
  Request 1 →
             Response 1 (takes 500 ms) ←
  Request 2 →
             Response 2 (takes 500 ms) ←
  Request 3 →
             Response 3 (takes 500 ms) ←
  
  Total: 1500 ms (sequential, pipelined doesn't help much)
  
  If Request 1 delayed (500 ms wait):
  Requests 2 and 3 stuck waiting
  "Head of line" (first in queue) blocks rest
  

  Problem 2: Limited Parallelism

  HTTP/1.1: 6-8 parallel connections (browser limit)
  
  Modern website: 100+ resources
  - 100 JavaScript files
  - 50 images
  - 20 CSS files
  - Fonts, videos, etc.
  
  Only 6 can download simultaneously
  Rest wait in queue
  
  Creating more connections wastes resources:
  - TCP handshake overhead
  - TLS negotiation overhead
  - Congestion window reset
  

  Problem 3: Header Compression Missing

  HTTP/1.1 headers: Plain text, often repeated
  
  Example request:
  GET /image1.jpg HTTP/1.1
  Host: www.example.com
  User-Agent: Mozilla/5.0...
  Accept-Language: en-US,en;q=0.9
  Accept-Encoding: gzip, deflate, br
  Cookie: session=abc123; user=john; ...
  
  Size: ~500 bytes
  
  Next request (same domain):
  GET /image2.jpg HTTP/1.1
  Host: www.example.com
  User-Agent: Mozilla/5.0...
  Accept-Language: en-US,en;q=0.9
  Accept-Encoding: gzip, deflate, br
  Cookie: session=abc123; user=john; ...
  
  Same 500 bytes again!
  90% of headers are identical
  Massive waste sending headers repeatedly
  

  Problem 4: Text-Based Overhead

  HTTP/1.1: Text-based parsing
  
  GET /index.html HTTP/1.1\r\n
  Host: example.com\r\n
  Connection: keep-alive\r\n
  \r\n
  
  Parser must:
  - Read character by character
  - Find line breaks (\r\n)
  - Parse key: value pairs
  - Handle whitespace variations
  - Error-prone
  
  Also: Humans can read it (debug), but wasteful
  

  HTTP/2 Solutions:

  Solution 1: Multiplexing

  HTTP/2: Multiple streams on single connection
  
  Request 1 (Stream 1) →     Frame 1
                             Frame 1
  Request 2 (Stream 3) →     Frame 2
                             Frame 1
  Request 3 (Stream 5) →     Frame 3
                             Frame 2
                             Frame 3
  ← Response 1 (Stream 1)
  ← Response 2 (Stream 3)
  ← Response 3 (Stream 5)
  
  Single connection, all streams active simultaneously
  No head-of-line blocking
  
  Timeline:
  Time 0:    Send req1, req2, req3 (all at once)
  Time 100:  Response 1 arrives (quick)
  Time 200:  Response 3 arrives (quick)
  Time 300:  Response 2 arrives (slow but others not blocked)
  
  Total: 300 ms (instead of 1500 ms with HTTP/1.1)
  

  Solution 2: Binary Framing

  HTTP/1.1: Text-based
  HTTP/2: Binary framing layer
  
  Each message split into frames:
  - Headers frame
  - Data frame (s)
  - Trailers frame
  
  Frame format:
  [Length: 3 bytes][Type: 1 byte][Flags: 1 byte]
  [Stream ID: 4 bytes][Payload: variable]
  
  Advantages:
  - Efficient parsing (binary, not text)
  - Fixed format, easier to implement
  - Multiplexable (each frame tagged with stream ID)
  - Can prioritize frames
  

  Solution 3: Header Compression (HPACK)

  HTTP/2: HPACK header compression
  
  First request:
  Host: www.example.com
  User-Agent: Mozilla/5.0...
  Accept: text/html
  
  Size: 500 bytes
  
  Second request (same domain):
  Only changed field: User-Agent
  
  Instead of resending all 500 bytes:
  Send: "Request 2 uses previous headers except User-Agent"
  Size: 50 bytes (10% of original!)
  
  How it works:
  - Maintain header table at client and server
  - Reference previous headers by index
  - Only transmit differences
  - Typical compression: 85-90% reduction
  

  Solution 4: Server Push

  HTTP/1.1:
  1. Browser requests index.html
  2. Server responds with index.html
  3. Browser parses, sees <link rel="stylesheet" href="style.css">
  4. Browser requests style.css
  5. Server responds
  
  Latency: 2 round-trips for html + css
  
  HTTP/2 Server Push:
  1. Browser requests index.html
  2. Server responds with index.html
  3. Server predicts client needs style.css
  4. Server proactively PUSH style.css
  5. Browser receives both in parallel
  
  Latency: 1 round-trip for html + css
  Browser doesn't need to wait for HTML before knowing about CSS
  

  Real-World Performance Improvement:

  Benchmark: Loading website with 100 resources
  
  HTTP/1.1 (6 parallel connections):
  - TCP handshakes: 6 × 50 ms = 300 ms
  - TLS handshakes: 6 × 100 ms = 600 ms
  - Header transmission: 100 × 500 bytes = 50 KB
  - Head-of-line blocking: Significant
  Total: ~3-5 seconds
  
  HTTP/2 (1 connection, multiplexing):
  - TCP handshake: 1 × 50 ms = 50 ms
  - TLS handshake: 1 × 100 ms = 100 ms
  - Header transmission: Compressed 85% = 7.5 KB
  - No head-of-line blocking
  - Binary efficient parsing
  Total: ~1-2 seconds
  
  Improvement: 2-3x faster
  

  Conclusion:

  HTTP/2.0 was motivated by HTTP/1.1’s inefficiencies: head-of-line blocking, limited parallelism (6-8 connections), repeated headers, and text-based overhead. HTTP/2 addressed these through multiplexing (many streams on one connection), binary framing (efficient parsing), header compression (HPACK), and server push (proactive delivery), resulting in 2-3x faster load times while reducing bandwidth usage.

• (n) Briefly describe the motivation for HTTP/3.0.

  View Answer

  In short: HTTP/3.0 was motivated by latency problems with TCP and TLS handshakes on high-latency or lossy networks. HTTP/3 replaces TCP with QUIC (UDP-based), which supports 0-RTT resumption, faster handshakes, connection migration, and per-stream congestion control to improve performance on mobile and unreliable networks.

  Elaboration:

  Problems with HTTP/2 (and TCP):

  Problem 1: TCP Handshake Overhead

  Every new HTTPS connection requires:
  
  1. TCP handshake (3-way):
     Client → SYN
     Server ← SYN-ACK
     Client → ACK
     Latency: 1 RTT
  
  2. TLS 1.2 handshake:
     Client → ClientHello
     Server ← ServerHello, Certificate, ...
     Client → ClientKeyExchange, ...
     Server ← Finished
     Client → Finished
     Latency: 2 RTT (minimum)
  
  Total: 3 RTT before data transmission
  On high-latency networks (100 ms RTT):
    3 × 100 ms = 300 ms just for handshakes!
  
  Mobile: Even worse (high latency, variable)
  

  Problem 2: Head-of-Line Blocking at TCP Layer

  HTTP/2 multiplexes over TCP, but TCP itself has HoL blocking:
  
  TCP guarantees ordered delivery:
  
  Packet 1 (Request A) sent at time T
  Packet 2 (Request B) sent at time T+10ms
  
  Network: Packet 1 dropped, Packet 2 arrives
  
  TCP must:
  1. Wait for Packet 1 retransmission
  2. Before delivering Packet 2 to application
  
  Even though Packet 2 arrived, application must wait
  HTTP/2 streams are blocked by TCP HoL blocking
  
  QUIC: Each stream has independent congestion control
  Dropped packet only affects its stream
  Other streams unaffected
  

  Problem 3: Connection Establishment Too Expensive

  HTTP/2 over TCP:
  - Users expect instant page load
  - Establish connection on first request
  - 3 RTT overhead unacceptable
  
  Mobile example:
  - RTT: 100 ms (common on 4G)
  - 3 RTT = 300 ms just for connection setup
  - Adds to total load time
  
  Reusing connection helps, but:
  - Network change (WiFi to cellular): Connection drops
  - Roaming between networks: Connection dies
  - Mobile users move frequently
  - Each new connection: 300 ms penalty
  

  Problem 4: Congestion Control Per Connection

  HTTP/2 scenario:
  
  Single TCP connection for multiple streams
  One packet loss → affects ALL streams
  
  Example:
  - Stream 1 (video): Can tolerate loss
  - Stream 2 (API call): Needs low latency
  
  Packet loss detected:
  TCP backs off (exponential backoff)
  Both streams slow down equally
  
  Inefficient: Video stream can wait, API can't
  

  HTTP/3 Solutions:

  Solution 1: QUIC Protocol (UDP-based)

  HTTP/3 uses QUIC instead of TCP
  
  QUIC = Quick UDP Internet Connection
  Runs on UDP (connectionless, fast)
  Implements reliability at QUIC layer (not TCP)
  
  Advantages:
  - No TCP handshake overhead
  - Faster connection setup
  - Connection migration support
  - Per-stream flow control
  - Per-stream congestion control
  

  Solution 2: 0-RTT Connection Establishment

  TLS 1.3 + QUIC enable 0-RTT:
  
  First connection:
  Client → Initial (with ClientHello data)
  Server ← Response + data
  Latency: 1 RTT (down from 3!)
  
  Resumed connection (within session ticket):
  Client → ClientHello from cache
  Server ← Data
  Client sends HTTP request with first packet!
  Latency: 0 RTT (literally sent with setup!)
  
  Example on 100 ms RTT:
  HTTP/2: 300 ms (3 RTT) setup
  HTTP/3: 0 ms (0 RTT) for resumed, 100 ms (1 RTT) for fresh
  

  Solution 3: Per-Stream Congestion Control

  HTTP/2 TCP problem:
  Single connection → single congestion window
  One dropped packet → all streams slow down
  
  HTTP/3 QUIC solution:
  Each stream manages its own congestion
  
  Scenario:
  - Stream 1 (video): Congestion window backing off
  - Stream 2 (API): Maintains own aggressive window
  - Streams don't interfere
  
  Video doesn't block API from sending
  Better performance for mixed traffic
  

  Solution 4: Connection Migration

  TCP problem:
  Connection identified by IP:port pair
  IP changes → TCP connection drops
  
  User scenario:
  1. Download starts on WiFi
  2. User walks out of WiFi range
  3. Device switches to cellular (IP changes)
  4. TCP connection drops
  5. Download restarts (wasted time, data)
  
  QUIC solution:
  Connections identified by Connection ID (not IP)
  IP change → Connection continues!
  
  User scenario with HTTP/3:
  1. Download starts on WiFi (QUIC connection)
  2. User walks out of range
  3. Device switches to cellular (IP changes)
  4. QUIC sends: "Still me, same Connection ID"
  5. Download resumes seamlessly
  6. No latency penalty, no data loss
  

  Solution 5: Faster Handshakes

  QUIC handshake (1 RTT minimum):
  
  Client → Initial packet
  Server ← Handshake packet + encrypted data
  Client → Acknowledgment
  
  Then: HTTP/3 request sent immediately
  
  Much faster than TCP (3 RTT) + TLS (2 RTT)
  

  Real-World Performance Impact:

  Scenario: Mobile user downloads webpage
  
  HTTP/1.1:
  - Network change: Connection drops
  - Must reconnect: TCP (1 RTT) + TLS (2 RTT) = 300 ms
  - Then: Re-download = slow
  
  HTTP/2:
  - Same problem: TCP + TLS overhead on reconnect
  - Still multiplexed, but same TCP latency issue
  
  HTTP/3:
  - Network change: QUIC migrates automatically
  - Connection continues: 0 RTT overhead
  - Download resumes instantly
  - Perception: Seamless, fast
  

  Comparison:

  Aspect	HTTP/2 (TCP)	HTTP/3 (QUIC)
  Initial handshake	3 RTT (TCP) + 2 RTT (TLS) = 5 RTT	1 RTT + 0-RTT resumption
  Head-of-line blocking	At TCP layer	Per-stream only
  Connection migration	Breaks	Seamless
  Mobile friendliness	Poor (reconnects drop)	Excellent (transparent migration)
  High-latency networks	500-1000 ms setup	100-200 ms setup

  Deployment Status:

  HTTP/3 adoption:
  - Chrome: Full support (2020+)
  - Firefox: Full support (2021+)
  - Safari: Full support (2022+)
  - Major sites: Google, Facebook, Cloudflare, etc.
  
  Benefits visible on:
  - Mobile networks (high RTT)
  - Network changes (WiFi → cellular)
  - High packet loss scenarios
  

  Conclusion:

  HTTP/3.0 was motivated by TCP’s overhead (3+ RTT handshakes, per-connection congestion control) and poor performance on mobile/unreliable networks. QUIC (UDP-based) provides 0-RTT resumption, connection migration without dropping, per-stream congestion control, and faster handshakes. This results in dramatically better performance on mobile, high-latency, and unstable networks where connection drops are common.

Problem 2: Dynamic Host Configuration Protocol

We discussed in class that a host’s IP address can either be configured manually, or by Dynamic Host Configuration Protocol (DHCP).

• (a) Describe the advantages and disadvantages of each approach.

  View Answer

  Manual IP configuration assigns fixed addresses, providing stability and predictability but suffering from poor scalability, configuration errors, and IP conflicts. DHCP automatically assigns IP addresses and network parameters, scales well, and reduces errors, but depends on a DHCP server and may result in changing IP addresses.

• (b) Describe how a host gets an IP address using DHCP.

  View Answer

  A host uses DHCP via the DORA process: it broadcasts DHCPDISCOVER, receives a DHCPOFFER, responds with DHCPREQUEST, and receives DHCPACK, after which it configures its network interface.

Problem 3: Video Streaming Protocol Selection

Consider an application where a camera at a highway is capturing video of the passing cars at 30 frames/second and sending the video stream to a remote video viewing station over the Internet. You are hired to design an application-layer protocol to solve this problem. Which transport-layer protocol, UDP or TCP, would you use for this application and why? Justify your answer.

View Answer

UDP is preferred for real-time video streaming because it is delay-sensitive and can tolerate packet loss. UDP avoids retransmissions, congestion control delays, and head-of-line blocking present in TCP, resulting in smoother playback.

Problem 4: DNS Recursive vs Iterative Queries

Consider a host H within qc.cuny.edu domain, whose name server is ns.qc.cuny.edu. Suppose that H tries to learn the IP address of the host ringding.cs.umd.edu. Assume that ns.qc.cuny.edu does not have the IP address of ringding.cs.umd.edu in its cache. Further assume that root DNS servers only know the authoritative name server for umd.edu domain.

• (a) Describe how the IP address of ringding.cs.umd.edu will be resolved assuming no DNS server implements recursive queries.

  View Answer

  With iterative queries, the local DNS server queries the root server, then the umd.edu server, then the cs.umd.edu server, finally obtaining the IP address of ringding.cs.umd.edu and returning it to the host.

• (b) Redo (a) assuming ALL DNS servers implement recursive queries.

  View Answer

  With recursive queries, the host sends one query to its local DNS server, which recursively contacts all necessary DNS servers and returns the final IP address.

Problem 5: DNS and HTTP Web Page Retrieval

Suppose within your Web browser you click on a link to obtain a Web page. Suppose that the IP address for the associated URL is not cached in your local host so that a DNS look-up is necessary to obtain the IP address. Suppose that $n$ DNS servers are visited before your host receives the IP address from DNS; the successive visits incur RTTs of $RTT_1, RTT_2, \ldots, RTT_n$. Let $RTT_0$ be the RTT between your local host and the Web server containing the Web page and let $B$ bits/sec be the sustained bandwidth.

• (a) Suppose that the Web page consists of a single object of size $D_1$ bits. Further suppose that the DNS queries are sent over UDP. How much time elapses from when the client clicks on the link until the client receives the object?

  View Answer

  Total time: $\sum_{i=1}^{n} RTT_i + 2RTT_0 + \frac{D_1}{B}$

• (b) Redo (a) assuming that the DNS queries are sent over TCP.

  View Answer

  Total time: $\sum_{i=1}^{n} 2RTT_i + 2RTT_0 + \frac{D_1}{B}$

• (c) Assume that the user clicks on a link within the just downloaded page and starts downloading a new web page of size $D_2$ bits residing at the same server. Assume this page also consists of a single object. How much time elapses from when the client clicks on the new link until the client receives the new object? Assume that the DNS uses UDP as in (a).

  View Answer

  Total time: $2RTT_0 + \frac{D_2}{B}$

• (d) Now assume that the web page to be downloaded in (c) has 6 other embedded objects each with size $D_3$. Assuming that the Web browser implements HTTP/1.0 with non-persistent connections and no parallel TCP connections, how much time elapses from when the client clicks on the new link until the client receives all objects?

  View Answer

  Total time: $(2RTT_0 + \frac{D_2}{B}) + 6(2RTT_0 + \frac{D_3}{B})$

• (e) Redo (d) assuming the Web browser implements 4-parallel TCP connections with non-persistent connections.

  View Answer

  Total time: $(2RTT_0 + \frac{D_2}{B}) + 4RTT_0 + \frac{6D_3}{B}$

• (f) Redo (d) assuming the Web client uses HTTP/1.1 with persistent connections (no pipelining).

  View Answer

  Total time: $(2RTT_0 + \frac{D_2}{B}) + 6(RTT_0 + \frac{D_3}{B})$

Problem 6: Instant Messaging System Architecture

Suppose you were to implement an instant message such as Yahoo messenger, which allows any number of users to exist in the system and establish instant messaging sessions among them.

• (a) Describe the architecture of your system (system components, protocol messages exchanged etc.) to enable users to dynamically learn each other’s current IP addresses and port numbers so that they can seamlessly start instant messaging sessions.

  View Answer

  Users register IP addresses and ports with a centralized directory server, which clients query to discover peers, while messages are exchanged peer-to-peer.

• (b) Suppose you were to allow users to have “buddy lists” and learn about the current communication status of their buddies. How would you extend your system to enable this feature?

  View Answer

  The directory server maintains buddy lists and presence information and notifies users of status changes.

Problem 7: Web Proxy Caching Performance

Assume that Queens College decided to use a Web Proxy, i.e., a Web cache. In this model, each Web browser is set up to send their requests to the Web proxy rather than sending the request directly to the actual Web server. Recall that a Web browser also maintains a local cache. Suppose a user accesses 100 objects one after the other using HTTP/1.0. The size of each object is 10000 bits. Assume that the sustained bandwidth between the user’s PC and the Web Proxy is 10Mbps and has an RTT of 1ms, and the sustained bandwidth between the Web proxy and a Web server is 1Mbps and has an RTT of 100ms.

• (a) What is the average object retrieval time in the absence of any cache hits?

  View Answer

  With no cache hits, the average object retrieval time is approximately 101 ms.

• (b) Assume that of all user requests, 20 percent is found in the Browser cache, half of the remaining user requests are satisfied from the Web Proxy cache, and the remaining requests make it up to the Web server. What’s the average object retrieval time now?

  View Answer

  With browser and proxy caching, the average retrieval time is approximately 40.8 ms.

Problem 8: Alert Notification Protocol Selection

Consider a network-attached burglar alarm which is programmed to notify the police when a burglar enters the house. Suppose that you are to use either HTTP or SMTP to send the notification message. How would you use each protocol to send the message? Which protocol makes more sense to use for this application?

View Answer

SMTP is preferable to HTTP for alarm notifications because it provides reliable store-and-forward delivery.

Problem 9: Email with MIME Attachments

Suppose you want to send an e-mail message M with 4 attachments, A1, A2, A3 and A4. Describe how your e-mail client, e.g., Outlook, would send this e-mail?

View Answer

The client constructs a MIME multipart message with Base64-encoded attachments and sends it using SMTP.

Problem 10: POP3 and IMAP Mail Protocols

What are POP3 and IMAP used for? What are the advantages of IMAP over POP3?

View Answer

POP3 downloads email locally, while IMAP keeps email on the server and supports synchronization and folders, making IMAP more flexible.